BWI Hackathon leverages superior GPU Power with NVIDIA H100 and L4

In this article, you will learn

• what the German armed forces uses artificial intelligence for,
• how rapidly prototypes can be developed in a hackathon setting
• and why cloud-based GPUs are critical for AI innovation.

From idea to prototype in just five days: at the joint Data Analytics Hackathon organized by BWI and the Bundeswehr, teams developed AI-based solutions for concrete Bundeswehr use cases, supported by GPU resources from the Open Telekom Cloud. 

At the end of November, BWI, the IT service provider of the German Armed Forces, hosted its seventh BWI Data Analytics Hackathon at the Aumühle site in Bonn. More than 100 participants, working both on site and remotely, formed 20 teams. Over the course of five days, they tackled four challenges defined by BWI and the Bundeswehr, each requiring varying degrees of AI deployment. The challenges were as follows:

1. Wargaming: Speech-to-text analysis to support military decision-making
2. OSINT: Automated situational awareness using social media, geospatial, and open data sources
3. Space Weather: Analysis of solar storms and magnetic pulses based on satellite data
4. Software Engineering: Development of an offline mapping solution comparable to Google Maps for use by the German Armed Forces

“Code freely, no regular dates, no meetings”: The right mindset is the most important ingredient of a successful hackathon. At the same time, creativity and compelling challenges alone are not sufficient. A hackathon also depends on a powerful, stable, and flexible technical platform that allows all teams to work in a focused and uninterrupted manner.

This is precisely how the Open Telekom Cloud, together with the T-Systems onsite Innovation Lab, actively supported this year’s BWI Hackathon (read more here). From November 24 to 28, participants were able to develop and test their ideas using a dedicated GPU environment. In total, 23 GPU instances equipped with NVIDIA L4 and H100 GPUs were deployed, enabling high-performance execution of a wide range of AI and big data workflows and use cases.

Robert Renz and Florian Krumbholz from the T-Systems onsite Innovation Lab were present throughout the event to review projects and prototypes firsthand. They supported participants by answering technical questions and acting as sparring partners for in-depth discussions and feedback. “It was great to see how quickly ideas became concrete, workable solutions,” they noted. “In the future, we will continue to be ready with sovereign infrastructure, a broad GPU portfolio, and the appropriate know-how to promote innovation with the necessary cloud resources.” 

Much like the hackathon challenges themselves, the T-Systems Innovation Lab delivers onsite projects using a minimum viable product (MVP) approach. Following a brief requirements analysis conducted jointly with the customer, an initial, executable prototype is delivered within a short timeframe. This prototype then serves as the foundation for further iterations, helping to make ideas tangible and actionable. These MVPs typically combine cloud technologies, IoT, and custom software development to create intelligent and innovative solutions.

The ECS compute flavors with NVIDIA L4 GPUs (pi5e) and NVIDIA H100 GPUs (p5s) are core components of the GPU portfolio of Open Telekom Cloud. The L4-based instances are optimized for inference, small to medium-sized models, and multimodal workloads, while the H100-based flavors deliver the compute performance required for highly demanding deep learning workloads and advanced fine-tuning scenarios. Together, the GPU flavors featuring NVIDIA L4 (pi5e) and H100 (p5s) underpin a wide range of AI use cases on the Open Telekom Cloud, from rapid prototyping to scalable production deployments. 

The hackathon demonstrated that the Open Telekom Cloud offers a state-of-the-art GPU portfolio capable of enabling advanced AI innovation. Organizations that intend to use GPU resources strategically over extended periods can benefit from discounted three-year packages to reduce overall costs. At the same time, the Open Telekom Cloud continues to expand its offerings, with new GPU options already on the horizon.

At the beginning of 2026, the Open Telekom Cloud will further expand its GPU portfolio with the new p5e flavor, featuring NVIDIA H100 GPUs connected via NVLink. This configuration combines two H100 Tensor Core GPUs using the high-speed NVLink interconnect, delivering substantially higher bandwidth and scalability than traditional PCIe-based setups. Direct GPU-to-GPU communication enables significantly faster and more efficient execution of AI models, simulations, and training processes. Workloads in generative AI, large language models (LLMs), and high-performance computing (HPC) benefit in particular from this architecture.

As outlined above, NVIDIA L4 GPUs are especially well suited for production-grade inference workloads. Complementing single high-performance GPUs, the Open Telekom Cloud is introducing a new architecture based on multiple L4 GPUs deployed in a high-availability (HA) cluster. This approach offers clear advantages in terms of scalability, stability, and load balancing.

By combining multiple L4 GPUs, inference workloads can be scaled horizontally, with requests dynamically distributed across instances to eliminate bottlenecks and ensure balanced utilization. In the event of a GPU or node failure, other cluster nodes automatically assume the workload, ensuring uninterrupted AI services in production environments.

This architecture is particularly well suited for latency-sensitive applications such as chatbots, recommendation systems, and real-time image analysis.